Skip to main content

AV Index v1

Methodology

This page describes how the live tables are calculated today. Every score rests on verified data only: regulator filings we re-ingest automatically, sourced company figures, and recent news. It is not a regulator score, a crash-rate ranking, or a Times Higher Education-style research assessment. Sector rank is the primary public number. Global AV is a cross-sector normalization, not “best robotaxi.”

Inclusion

Companies on the boards have public status active, acquired, or subsidiary, plus is_verified and audit verdict verified. The directory on /companies is the same verified set. Unverified or pending records are not on the public boards. Being on a board is not the same as holding a rank: see “Who gets a rank” below.

Pillars and what counts as verified

A sector score is a weighted sum of five pillars, each 0–100. A pillar counts only when verified data backs it:

  • Safety — California DMV autonomous-vehicle mileage and disengagement reports and NHTSA Standing General Order crash reports, re-ingested every week. Miles per disengagement on a log scale (at least 10,000 reported miles); a company that only appears in NHTSA crash reports starts at 60; the share of crashes with a serious or fatal injury alleged lowers the score by up to 30 points. Data older than two years is ignored. This is not a Euro NCAP grade or a crash rate per mile.
  • Deployment — city counts or miles that carry a corroborated source, or autonomous miles reported to a regulator. Supervised miles are not treated as paid driverless service.
  • Capital — market capitalisation, funding or revenue with a source: SEC filings (annual revenue and, for US-listed companies, the public float on the annual report’s cover, the market value of the shares held by non-affiliates, which stands in for market capitalisation), Wikidata, a matched company page, or an accepted reader correction. Figures from filings and Wikidata refresh every week.
  • Momentum — how many approved news articles of the last 30 days name the company, counted as whole words (a name inside another word does not count, and a name that is also an everyday word counts only when capitalised inside a sentence). With fewer than two articles it is not counted. The model’s sentiment score and how recently our own record was edited play no part.
  • Strategy — no verified source yet, so it does not separate companies. Outlook and AV-exposure ratings shown on the site are model estimates for context and never enter a score.

A pillar without verified data is set to the median of the verified values on the same board (50 when nobody has data), so a gap neither sinks a company as a zero would nor lifts it. Next to each score the board shows how much of it is verified: the share of the sector weights that verified data backs. Dashes in the breakdown mark the pillars that sit at the board median.

SectorSafetyDeployStrategyCapitalMomentum
Robotaxi Ops35%30%15%10%10%
Automotive OEMs20%20%25%25%10%
Autonomous Trucking30%30%15%15%10%
Hardware & Compute10%10%40%25%15%
Software & AI10%10%40%25%15%
Last-Mile Delivery30%30%15%15%10%
Industrial AVs25%30%20%15%10%

Who gets a rank

The rankings page offers two levels of strictness. A level decides who earns a rank, never a company’s score: a pillar without verified data always sits at the board median. Standard is the default.

  • Standard: on the operator boards (Robotaxi Ops, Autonomous Trucking, Last-Mile Delivery) a company is ranked once at least one verified safety or deployment figure exists; capital and news alone do not make an operator. On the other boards a company is ranked once at least 20% of the sector weights is backed by verified data.
  • Strict: a verified safety or deployment figure is needed on every board, plus at least 40% of the weights verified (in practice two verified pillars or more). Boards whose companies publish no such data, such as hardware and software, then rank nobody.
  • At every level a board orders companies only when at least three qualify. Otherwise its companies are listed without a rank.
  • A company without enough verified data stays listed under the ranking, unranked, with its verified share. It is not on the global board. It ranks automatically once its data arrives: the sources refresh every week.

Business lines

Every company has one home board (its primary sector; overrides: Tesla, Hyundai and JLR as OEMs, Mobileye as Software, Nuro as Last-Mile, Aurora and Waabi as Trucking). Robotaxi Ops also lists companies from other home boards that are in the robotaxi business, such as Tesla with its robotaxi service and Cybercab. Membership depends on being in the business, not on how mature the service is: maturity only moves the score, through the safety and deployment data. Suppliers that do not run a service (Mobileye, 42dot) and trucking or delivery developers stay off Robotaxi Ops. Stage (Commercial, Pilot, Testing) is a view on the operator boards, not a rank gate.

Roles

Archetypes are derived from primary sector: operator (robotaxi, trucking, last-mile), enabler (hardware, software), oem, or industrial specialist. The homepage highlights Robotaxi Ops. Global AV is a separate board.

Global AV rank

Global score is the company’s score on its home board, multiplied by a role factor so operators typically lead the cross-sector board (operator 1.00, OEM 0.92, industrial specialist 0.82, enabler 0.78). Only companies ranked on their home board, at the strictness level chosen, appear. AV exposure and capital estimates no longer enter the global score because they are not verified. Cite the sector table unless you mean this cross-sector normalization.

Official records (not scored)

A company page can show a card of official records: a California DMV permit, the CPUC quarterly report of its passenger service, the research and development spending in its SEC annual report, federal contract awards, NHTSA recalls and investigations filed under its brands, and Federal Register notices that name it. Each record states where it comes from and the date the source gives, and links to it. They are context for the reader and never enter a score or decide who is ranked: a permit shows that a company tests, a recall count depends on how many vehicles a brand sells, and a federal contract says who the company sells to, not how well it drives. Contract awards are shown only for recipients checked by hand, and Federal Register notices only when their title names the company. The sources refresh every week.

Limitations

  • Verified safety data comes from California DMV and NHTSA only. A company that reports elsewhere, or not at all, has no verified safety yet. A shared 60 for crash-report-only companies means little until they also report miles.
  • Hardware & Compute and Software & AI currently differ mainly by capital and news momentum: strategy, 40% of their weight, has no verified source yet.
  • Coverage is uneven. Many smaller or non-US companies have little verified data and are listed unranked until a source covers them.
  • Capital uses the first sourced figure among market capitalisation, funding raised and annual revenue. The public float is a lower bound of the market capitalisation (it leaves out shares held by insiders, so it is far below it for a company that a parent controls). Funding taken from SEC Form D filings, the notices a private company files for its US offerings, is a lower bound as well: rounds sold abroad, under another exemption or by a parent are not in it, and the company page says “At least”. Companies listed outside the US and private companies have a capital figure only when a page states their market capitalisation or funding; until then the pillar sits at the board median.
  • The regulation page is an Index reading of public materials, grouped by country and common framework (US federal, EU type-approval). US states such as Arizona are nested under the United States, not treated as sovereign countries. It is not a live CA DMV/NHTSA feed or legal advice.
  • News volume is not quality. Adjacent or duplicate headlines can still affect momentum.
  • Market and funding figures on company pages and Market Intelligence are company-record estimates (and, on Market, AV-adjusted). They are not a cited industry total.
  • Deployment miles and city counts are incomplete for many firms, especially outside the US.
  • This is not an annual 15-name editorial list. The full verified set remains in the directory.

See the live tables on /rankings?sector=Robotaxi+Ops. Each table footer shows when that snapshot was calculated.